[RL25]: A Register Approach to Modal (Non-)Concord in English #
Modal concord (MC) doubles a modal verb and a modal adverb of the same force
(may possibly, must certainly) versus the single modal (SM, may / must).
[RL25] asks whether MC differs from SM in meaning and social
perception, and whether that difference is register-sensitive (varies with
situational context). Its Experiment 1 is the no-context study [LR25]
(formalized in Studies/LiuRotter2025); Experiment 2, formalized here, adds a
CONTEXT factor — interlocutor relation, close vs distant — in a 2×2×2
(NUMBER × FORCE × CONTEXT) design (306 participants, Prolific).
Two analyses make precise, divergent predictions ([RL25] (3a)/(3b)):
- concord ([Zei07]): MC ≡ SM, so doubling has no commitment effect, for any force.
- modal spread ([GM18]): for universal modals doubling strengthens commitment (must certainly > must); for existential modals it merely maintains the default (may possibly = may).
They differ only for universal modals. Experiment 2 replicates Experiment 1: the necessity strengthening (which adjudicates for spread), the possibility weakening (a residual neither analysis predicts, [RL25] §4.1), the confidence crossover, the warmth penalty, and lower grammaticality/appropriateness for MC. Crucially (RQ4) no NUMBER × CONTEXT interaction appears for any measure: MC's effect is not register-sensitive to interlocutor relation.
Main definitions #
concordPred/spreadPred— the two analyses asModalForce → SignType.RegisterSensitive— a concord effect whose sign differs between contexts.observedShift— the MC − SM cell-mean shift fromData.Examples.RotterLiu2025.
Main results #
analyses_diverge_only_on_necessity— concord and spread agree on possibility, differ on necessity: the single locus the data can adjudicate.necessity_adjudicates— the necessity strengthening matches spread, not concord.possibility_residual— the possibility weakening matches neither analysis.context_main_effect_preserves_sign/not_registerSensitive_of_main_effect— an additive CONTEXT main effect cancels in MC − SM, so it yields no NUMBER × CONTEXT interaction: the structural reason for the RQ4 null.possibility_observed_vs_predicted— the observed weakening ([LR25]'sspreadEffect, −1) versus the spread prediction (0) is exactly the residual.
Implementation notes #
Reuses the sign-prediction machinery of Studies/LiuRotter2025
(ShiftObservation, cellMean, forceKey, the rival accounts). Cell means live
in Data.Examples.RotterLiu2025 (×100, on the 1–7 scale); they do not reduce in
the kernel, so concrete shifts are computed via #eval while the
kernel-checkable content is each analysis's systematic prediction.
The two analyses (precise predictions) #
The concord analysis ([Zei07]): MC is truth-conditionally
equivalent to SM, so doubling has no commitment effect — the force-blind null
LiuRotter2025.vacuityEffect.
Instances For
The two analyses agree for existential modals (no change) and differ only for universal modals — the single locus where the data can adjudicate ([RL25] (3a) vs (3b)).
Predicting against the data #
Each cell of Data.Examples.RotterLiu2025 carries the Experiment 2 means
(×100) for one CONTEXT × FORCE × NUMBER combination. An analysis predicts the
sign of the concord shift MC − SM per force; the observed sign is read off the
paired cell means.
The cell with the given context / force / number paperFeatures.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The observed MC − SM shift for measure in context under force.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The universal case adjudicates: must certainly > must carries the strengthening sign the spread analysis predicts and the concord analysis does not.
The existential weakening is a residual: may possibly < may carries a sign that neither analysis predicts — both predict maintenance ([RL25] §4.1).
Register (in)sensitivity — RQ4 #
The headline null result: no NUMBER × CONTEXT interaction. Structurally, a CONTEXT main effect shifts the MC and SM means by the same amount, so it cancels in the concord contrast MC − SM and cannot change its sign.
A concord effect is register-sensitive (for this CONTEXT parameter) when its shift differs in sign between the close and distant contexts — a NUMBER × CONTEXT interaction at the level of sign.
Equations
- RotterLiu2025.RegisterSensitive close distant = (close.observedSign ≠ distant.observedSign)
Instances For
A CONTEXT main effect (the same additive shift c on MC and SM) leaves the
concord shift's sign unchanged: it cancels in MC − SM.
RQ4 ([RL25]): when contexts differ only by a main effect, the concord shift is not register-sensitive — close and distant carry the same sign, so no NUMBER × CONTEXT interaction arises.
Relationship to Experiment 1 ([LR25]) #
The replication is exact at the level of accounts on the necessity side, and the
one mismatch on the possibility side is precisely the residual: [LR25]'s
spreadEffect records the observed weakening (−1), whereas the spread
analysis predicts maintenance (0).
On necessity, Experiment 2's strengthening matches the spread prediction and the crossover sign [LR25] reports for Experiment 1.
On possibility, [LR25]'s observed spreadEffect (−1) diverges
from the spread analysis's prediction (0): the gap is the residual that
possibility_residual isolates.
The observed shifts #
The #evals exhibit Experiment 2's findings (means do not reduce in the kernel,
so these are computed). The commitment and confidence crossovers and the warmth
and appropriateness penalties hold in both contexts, and the commitment sign is
context-invariant per force — the RQ4 null in the data.